Training method of active tuberculosis detection model, detection device and equipment

Abstract

The invention relates to the technical field of pulmonary tuberculosis image classification methods, in particular to a training method of an active pulmonary tuberculosis detection model and a detection device and equipment.The method comprises the steps that front-position CXR data and lateral-position CXR data of a to-be-detected object are obtained, the CXR data in a preset proportion are marked, the marked CXR data form a first training data set, and the first training data set is used for training the to-be-detected object; the rest forms a second training data set; training a self-supervised model based on the second training data set to determine an encoder module, the encoder module being used for extracting a feature vector of an input image; training a supervised learning model based on the first training data set to generate the active tuberculosis detection model, the supervised learning model comprising a feature extraction module and a prediction module, the feature extraction module having the same network structure as the encoder module and being initialized by the encoder module; the accuracy of the active tuberculosis automatic screening method can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pulmonary tuberculosis image classification methods, and in particular to a training method and a detection device and equipment for an active pulmonary tuberculosis detection model. Background Art

[0002] Chest X-ray (CXR) imaging is a non-invasive medical imaging method for clinical screening of tuberculosis. At present, CXR-based tuberculosis screening mainly relies on clinicians to read the film to determine the imaging characteristics of the lesions, and the accuracy of identification is highly dependent on the doctor's subjective experience and image quality. At the same time, it is becoming more and more popular to use supervised learning-based neural network models to perform automated CXR imaging for tuberculosis screening. This method first annotates the CXR image data to generate a training data set, trains the deep learning neural network on the training data set, and generates a tuberculosis screening model. Finally, the tuberculosis screening model is used to detect active tuberculosis in the input CXR image data.

[0003] However, the current neural network-based automatic tuberculosis screening methods often only use frontal or lateral CXR data, but do not use data from both angles to analyze the characteristics of tuberculosis lesions, failing to give full play to the synergistic advantages of data; at the same time, medical image annotation is expensive, and tuberculosis lesions may appear in a variety of forms on CXR, such as nodules, infiltrations, cavities, etc., and the size, morphology and distribution of lesions may vary greatly. The cost of annotating massive CXR data based on the above-mentioned forms to form a large data training data set is almost unacceptable. However, the prediction accuracy of the tuberculosis screening model trained based on a small data training data set is not high. Summary of the invention

[0004] In view of this, the purpose of the embodiments of the present invention is to provide a CXR-based active tuberculosis detection method, device and equipment to solve the problem of low accuracy of the existing neural network-based automatic screening method for active tuberculosis.

[0005] In a first aspect, an embodiment of the present invention provides a training method for an active tuberculosis detection model, comprising: Acquire the frontal and lateral CXR data of the object to be detected, annotate a preset proportion of the CXR data, the annotated CXR data constitute a first training data set, and the remaining CXR data constitute a second training data set; Training a self-supervised model based on the second training data set to determine an encoder module, wherein the self-supervised model includes an encoder module, and the encoder module is used to extract a feature vector of an input image; Construct a supervised learning model, train the supervised learning model based on the first training data set, and generate the active pulmonary tuberculosis detection model, wherein the supervised learning model includes a feature extraction module and a prediction module, the feature extraction module has the same network structure as the encoder module and is initialized by the encoder module.

[0006] Optionally, performing self-supervised learning based on the second training data set to determine the encoder module includes: Based on the second training data set, self-supervised learning is performed on the encoder-decoder structure to generate an image reconstruction model and determine the weight of the encoder module, wherein the decoder module is used to reconstruct the image based on the feature vector.

[0007] Optionally, the encoder module comprises multiple attention modules to extract feature vectors of the input image from multiple scales.

[0008] Optionally, the image reconstruction model comprises at least two encoder-decoder branches, respectively used for processing anteroposterior CXR data and lateral CXR data.

[0009] Optionally, the feature extraction module is initialized by the encoder module including: The image reconstruction model includes two encoders, and the weights of the two encoders are fused according to a preset rule to generate a fusion weight; The fusion weight is used as the initial value of the feature extraction module weight.

[0010] Optionally, the supervised learning model also includes a feature mining module, which is used to perform feature fusion on the multi-scale feature vector output by the feature extraction module based on preset rules and output the fused features to the prediction module for active tuberculosis prediction.

[0011] Optionally, the performing feature fusion on the multi-scale feature vector output by the feature extraction module based on a preset rule includes: Selecting a feature vector of a target size from the multi-scale feature vectors based on a preset rule; The selected feature vectors are subjected to global feature mining based on multi-head attention and local feature mining based on a residual network to obtain enhanced features, and feature fusion is performed based on the enhanced features.

[0012] Optionally, the training the supervised learning model based on the first training data set includes: Performing data enhancement on the data in the first training data set; Using the fusion weight as the initial weight value of the feature extraction module, freezing the feature extraction module, and training the remaining parameters of the supervised learning model except the feature extraction module based on the first training data set after data enhancement; After the training target is reached, the feature extraction module is unfrozen, and then all parameters of the supervised learning model are trained based on the first training data set after data enhancement.

[0013] In a second aspect, an embodiment of the present invention provides a method for detecting active pulmonary tuberculosis, comprising: Acquire CXR data of the object to be detected; The CXR data is input into the active pulmonary tuberculosis detection model trained according to the method described in the first aspect of the present invention, and the active pulmonary tuberculosis detection result is output.

[0014] In a third aspect, an embodiment of the present invention provides an active tuberculosis detection device, comprising: An acquisition module, used for acquiring CXR data of the object to be detected; The detection module is used to input the CXR data into the active pulmonary tuberculosis detection model trained according to the method described in the first aspect of the present invention, and output the active pulmonary tuberculosis detection result.

[0015] In a third aspect, an embodiment of the present invention provides an active tuberculosis detection device, comprising a memory and a processor. The memory is used to store computer programs; The processor is used to implement the steps of the following method when executing the computer program: Acquire CXR data of the object to be detected; The CXR data is input into the active pulmonary tuberculosis detection model trained according to the method described in the first aspect of the present invention, and the active pulmonary tuberculosis detection result is output.

[0016] The embodiments of the present invention include the following beneficial effects: the present embodiment performs active tuberculosis detection based on medical image data of CXR active tuberculosis detection in frontal and lateral positions, and because the synergistic effect of CXR data acquired at different angles of the target part is utilized, the accuracy of chest lesion positioning and classification can be improved; at the same time, the present disclosure uses unlabeled samples to pre-extract deep features of CXR images in a self-supervised manner, and then fuses the pre-trained encoder through learnable parameters, and shares the fused weights with the encoder of the supervised learning task, thereby achieving the purpose of using fewer labeled samples to improve detection accuracy, wherein local attention (cross-channel attention) and global attention (multi-head convolution attention) are introduced at the same time to fully mine local and global semantic information, and in the global attention, 3x3 depth-separable convolution is used to optimize the features generated after multi-head attention splicing to reduce information loss, and multi-scale information is used to enhance the accuracy of detection target positioning and classification, and when performing multi-scale feature fusion, only the last two feature maps are used for combination, which does not cause feature redundancy and excessive computational cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0018] Figure 1 A flowchart of a method for training an active pulmonary tuberculosis detection model provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of the structure of the training network of the active tuberculosis detection model; Figure 3 It is a schematic diagram of the structure of the encoder in the self-supervised model; Figure 4 It is a schematic diagram of the structure of the MD module in the supervision model; Figure 5 It is a schematic diagram of the structure of the MAD module in the supervision model; Figure 6 It is a schematic diagram of multi-scale feature fusion in the supervised model; Figure 7 is a schematic diagram of an active pulmonary tuberculosis detection device according to some embodiments of the present invention; Figure 8 is a schematic diagram of an active tuberculosis detection device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0020] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0022] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0023] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0024] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0025] Figure 1 A flowchart of a training method for an active tuberculosis detection model provided in an embodiment of the present invention is applied to a model training server, and the method comprises the following steps: S110, acquiring frontal and lateral CXR data of the object to be detected, annotating a preset proportion of the CXR data, the annotated CXR data forming a first training data set, and the remaining CXR data forming a second training data set.

[0026] The frontal and lateral CXR data of the object to be detected are obtained to form a training data set. Specifically, the collected data is converted into PNG format, and the DICOM format data is converted into PNG format. The CXR image is cropped using the center to obtain a CXR image with a final size of 1024X1024X3. The data set is annotated as the input sample X={X1,X2,........,XN1} of the supervised task, and the remaining unlabeled data is used as the input sample X~={X1,X2,........,XN2} of the self-supervised task.

[0027] S120: Training a self-supervised model based on the second training data set to determine an encoder module, wherein the self-supervised model includes an encoder module, and the encoder module is used to extract a feature vector of an input image.

[0028] First, mask the input data.

[0029] Figure 2 This is a schematic diagram of the structure of the training network for the active tuberculosis detection model. Figure 2 As shown, the contrast of the image is enhanced by contrast-constrained adaptive histogram equalization (CLAHE) for the sample X~={X1,X2,........,XN2} in the self-supervised stage. The contrast of the image is converted into a binary image to highlight the information, and then it is inverted into pixel values ​​for further processing. The inverted binary image is used to retrieve the lung cavity contour through the findContours function provided by OpenCV. n points are randomly sampled in the candidate lung cavity area as the center coordinate points to generate a square mask of size 96x96 (the pixel value in the area is 0, and the pixels in the remaining part are equal to 1). The mask image is obtained by superimposing the original image by element-wise multiplication, thereby creating a mask image set M.

[0030] Then, a self-supervised image reconstruction model is constructed.

[0031] The self-supervised image reconstruction model is an encoder-decoder structure. The encoder structure is as follows Figure 3 As shown in the figure, the input image is smoothed by 3x3 convolution, and then captured by 5 EConv modules with similar structures. Each EConv is followed by feature map compression by Maxpooling to generate feature maps of different scales, with sizes of 512x512, 256x256, 128x128, 64x64, and 32x32 respectively.

[0032] EConv consists of 3x3 convolution and EMA modules. The 3x3 convolution smoothes the input image. The EMA module is an efficient multi-scale attention module. The output result is superimposed with the original image and then batch normalized.

[0033] The structure of the decoding part is similar to that of the encoding part, and the loss of a single image reconstruction is expressed as Loss = α·Lrec+β·adversarial loss Ladv.

[0034] , , in, Represents the masked area, x represents the input image, D(x) represents the variance, E(x) represents the expectation, and F(x) represents the generated image.

[0035] Finally, weight fusion is performed.

[0036] refer to Figure 1 , the weights Wz and Wc of the two self-supervised encoders are fused, including but not limited to the following fusion methods, and one or more combinations thereof may be selected.

[0037] 1. Average weight: Average the weight matrices of the two encoders on the same layer, expressed as Ws = (Wz + Wc) / 2.

[0038] 2 Weighted average: The weight matrices of the encoders from the frontal and lateral image reconstruction models are combined by weighted average through learnable parameters θz and θc, denoted as Ws = θz·Wz + θc·Wc.

[0039] 3 Product: Multiply the weight matrices and output a new matrix with unchanged dimension, expressed as Ws=Wz·Wc.

[0040] 4 Convolutional fusion: Wz and Wc pass through different convolutional blocks respectively to adaptively capture information of different scales.

[0041] 5 Attention fusion: Dynamically adjust the importance between Wz and Wc by learning a set of weights.

[0042] 6 Nonlinear fusion: Wz and Wc are respectively subjected to different nonlinear activation functions, such as ReLU, Sigmoid, Tanh and other functions, which make the features richer from a nonlinear perspective and are conducive to obtaining better feature fusion effects.

[0043] S130, constructing a supervised learning model, training the supervised learning model based on the first training data set, and generating the active pulmonary tuberculosis detection model, wherein the supervised learning model includes a feature extraction module and a prediction module, the feature extraction module has the same network structure as the encoder module and is initialized by the encoder module.

[0044] First, build a supervised learning model The structure of the feature extraction part of the supervised learning model is the same as the encoder structure of the self-supervised part, and the weights of the combination of the two self-supervised models are shared to achieve the purpose of pre-training.

[0045] Weight sharing refers to using the fused self-supervised encoder weights as the initial weights for the feature extraction part of the supervised model. The structure of the feature extraction part of the supervised model is consistent with the encoder structure of the self-supervised model, so these weights can be loaded directly.

[0046] The multi-scale feature maps extracted by EConv are smoothed through 5x5 convolution, BatchNormalization and ReLU. The deepest feature map is selected and passed through the MD module and MAD module for more detailed global and local feature mining. The two feature maps are spliced ​​and then spliced ​​with the feature map captured by the previous layer. They are combined only twice. The final output feature maps are connected to the classification head and the coordinate regression head respectively.

[0047] Figure 4 It is a schematic diagram of the structure of the MD module in the supervision model. Figure 4 As shown in the figure, the MD module is mainly composed of two residual modules. First, multi-head attention is used for global attention. The feature map size here is 32x32 and is not decomposed into patches. Then, linear projection is performed through two deep convolutions, which can retain the spatial position information of pixels and does not require position embedding.

[0048] Figure 5 It is a schematic diagram of the structure of the MAD module in the supervision model. Figure 5 As shown in the figure, the MAD module is mainly composed of two residual modules, namely a deep convolution-based module and a channel MLP-based module, where the input of the MLP-based module is the output of the deep convolution-based module. Both blocks are followed by channel scaling operations and DropPath operations to improve feature generalization and robustness capabilities.

[0049] The multi-scale feature fusion extracted by EConv uses only the last two feature maps for combination, such as Figure 6 Therefore, multi-scale feature fusion does not cause feature redundancy and high computational cost, but can use multi-scale information to enhance the accuracy of target positioning and classification.

[0050] Finally, the model is trained.

[0051] Data enhancement methods are used for all samples, including horizontal flipping, vertical flipping, random translation, adding Gaussian noise, etc., to help the model better generalize and adapt to various changing images.

[0052] First, the self-supervised model is trained. Then, the weights of the combined encoders of the two self-supervised models are used as the initial weight values ​​of the encoder of the supervised model. Next, the encoder of the supervised model is frozen, the model parameters are fine-tuned, the latter part of the encoder is trained separately, and then all the parameters of the model are trained. Finally, the image is input for active tuberculosis detection.

[0053] Some embodiments of the present invention provide an active pulmonary tuberculosis detection method, which is applied to an active pulmonary tuberculosis detection server, comprising: Get the CXR data of the object to be detected and input the CXR data into Figure 1 The active pulmonary tuberculosis detection model trained by the method described in S110-S130 outputs the active pulmonary tuberculosis detection result.

[0054] Figure 7 is a schematic diagram of an active tuberculosis detection device according to some embodiments of the present invention. Figure 7 As shown, the active pulmonary tuberculosis detection device includes an acquisition module 710 and a detection module 720. In some embodiments of the present disclosure, the active pulmonary tuberculosis detection function is performed by an active pulmonary tuberculosis detection server. An acquisition module 710 is used to acquire CXR data of the object to be detected; Detection module 720, used to input the CXR data into Figure 1 The active pulmonary tuberculosis detection model trained by the method described in S110-S130 outputs the active pulmonary tuberculosis detection result.

[0055] One embodiment of the present invention provides an active tuberculosis detection device. Figure 8 As shown, the active tuberculosis detection device includes a memory 820 and a processor 810, wherein the memory 820 is used to store a computer program; and the processor 810 is used to implement when executing the computer program. Figure 1 The method described in S110-S130.

[0056] In summary, the crop identification method, device and equipment provided by each embodiment of the present invention detects active pulmonary tuberculosis based on medical image data of CXR active pulmonary tuberculosis detection in frontal and lateral positions. Because the synergistic effect of CXR data obtained at different angles of the target part is utilized, the accuracy of chest lesion positioning and classification can be improved; at the same time, the present invention uses unlabeled samples to pre-extract deep features of CXR images in a self-supervised manner, and then fuses the pre-trained encoder through learnable parameters, and shares the fused weights with the encoder of the supervised learning task. Parameters, thereby achieving the purpose of using fewer labeled samples to improve detection accuracy, wherein local attention (cross-channel attention) and global attention (multi-head convolution attention) are introduced at the same time to fully mine local and global semantic information, and in the global attention, 3x3 depth-separable convolution is used to optimize the features generated after multi-head attention splicing to reduce information loss, and multi-scale information is used to enhance the accuracy of detection target positioning and classification. When performing multi-scale feature fusion, only the last two feature maps are used for combination, and the problem of feature redundancy and excessive computational cost will not be caused.

[0057] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0058] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0059] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0060] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0061] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0062] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0063] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0064] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0065] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A training method for an active pulmonary tuberculosis detection model, characterized in that: include: Acquire the frontal and lateral CXR data of the object to be detected, annotate a preset proportion of the CXR data, the annotated CXR data constitute a first training data set, and the remaining CXR data constitute a second training data set; Training a self-supervised model based on the second training data set to determine an encoder module, wherein the self-supervised model includes an encoder module, and the encoder module is used to extract a feature vector of an input image; Construct a supervised learning model, train the supervised learning model based on the first training data set, and generate the active pulmonary tuberculosis detection model, wherein the supervised learning model includes a feature extraction module and a prediction module, the feature extraction module has the same network structure as the encoder module and is initialized by the encoder module.

2. The method according to claim 1, characterized in that: The performing self-supervised learning based on the second training data set to determine the encoder module comprises: Based on the second training data set, self-supervised learning is performed on the encoder-decoder structure to generate an image reconstruction model and determine the weight of the encoder module, wherein the decoder module is used to reconstruct the image based on the feature vector.

3. The method according to claim 2, characterized in that: The encoder module includes multiple attention modules to extract feature vectors of the input image from multiple scales.

4. The method according to claim 3, characterized in that: The image reconstruction model includes at least two encoder-decoder branches, which are used to process the anteroposterior CXR data and the lateral CXR data respectively.

5. The method according to claim 4, characterized in that: The feature extraction module is initialized by the encoder module and includes: The image reconstruction model includes two encoders, and the weights of the two encoders are fused according to a preset rule to generate a fusion weight; The fusion weight is used as the initial value of the feature extraction module weight.

6. The method according to claim 5, characterized in that: The supervised learning model also includes a feature mining module, which is used to perform feature fusion on the multi-scale feature vector output by the feature extraction module based on preset rules and output the fused features to the prediction module for active pulmonary tuberculosis prediction.

7. The method according to claim 6, characterized in that: The performing feature fusion on the multi-scale feature vector output by the feature extraction module based on a preset rule comprises: Selecting a feature vector of a target size from the multi-scale feature vectors based on a preset rule; The selected feature vectors are subjected to global feature mining based on multi-head attention and local feature mining based on a residual network to obtain enhanced features, and feature fusion is performed based on the enhanced features.

8. The method according to claim 7, characterized in that: The training of the supervised learning model based on the first training data set includes: Performing data enhancement on the data in the first training data set; Using the fusion weight as the initial weight value of the feature extraction module, freezing the feature extraction module, and training the remaining parameters of the supervised learning model except the feature extraction module based on the first training data set after data enhancement; After the training target is reached, the feature extraction module is unfrozen, and then all parameters of the supervised learning model are trained based on the first training data set after data enhancement.

9. An active pulmonary tuberculosis detection device, characterized in that: include: An acquisition module, used for acquiring CXR data of the object to be detected; The detection module is used to input the CXR data into the active pulmonary tuberculosis detection model trained according to the method according to any one of claims 1-7, and output the active pulmonary tuberculosis detection result.

10. An active tuberculosis detection device, characterized in that: including memory and processor, The memory is used to store computer programs; The processor is used to implement the steps of the following method when executing the computer program: Acquire CXR data of the object to be detected; The CXR data is input into the active pulmonary tuberculosis detection model trained by the method according to any one of claims 1 to 8, and the active pulmonary tuberculosis detection result is output.

Images